Using optimal subset regression to identify factors associated with insulin resistance and construct predictive models in the US adult population

BACKGROUND: In recent decades, with the development of the global economy and the improvement of living standards, insulin resistance (IR) has become a common phenomenon. Current studies have shown that IR varies between races. Therefore, it is necessary to develop individual prediction models for each country. The purpose of this study was to develop a predictive model of IR applicable to the US population. METHOD: In total, 11 cycles of data from the NHANES database were selected for this study. Of these, participants from 1999 to 2010 (n = 14931) were used to establish the model, and participants from 2011 to 2020 (n = 13,646) were used to validate the model. Univariate and multivariable logistic regression was used to analyze the factors associated with IR. Optimal subset regression was used to filter the best modeling variables. ROC curves, calibration curves, and decision curve analysis were used to determine the strengths and weaknesses of the model. RESULTS: After screening the variables by optimal subset regression, variables with covariance were excluded, and a total of seven factors (including HDL, LDL, ALB, GLB, GLU, BMI, and waist) were finally included to establish the prediction model. The AUCs were 0.851 and 0.857 in the training and validation sets, respectively, and the Brier value of the calibration curve was 0.153. CONCLUSION: The optimal subset predictive model proposed in this study has a great performance in predicting IR, and the decision curve analysis shows that it has a high net clinical benefit, which can help clinicians and epidemiologists easily detect IR and take appropriate interventions as early as possible.